MQX Memory fragmentation

[jpfitz]

I have some questions regarding memory allocation and freeing with MQX that I hope someone can help with.  Our current application approach is using a lot of malloc and free calls, both of which we have pointed at the corresponding MQX memory functions for system memory (_mem_alloc() and _mem_free() ).  How does MQX address memory fragmentation and how well does MQX will keep track of a large number of freed memory holes for reuse on future malloc calls.  My understanding is that MQX will track the freed memory areas, and thier size, and use them for future allocations.  My questions are:

1. Is there a limit to the number of memory areas MQX will track? 
2. Does MQX look for a “first fit” or “best fit”? 
   i.e. after freeing 60 bytes, if a malloc  needs 10, will it take 10 out of the 60 byte freed area if it is the first memory area checked, or will it search for a better fitting memory hole?
3. Does MQX combine adjacent freed memory holes when freeing memory?  i
   .e. in the example from 2. If the 10 bytes are taken from the freed 60 bytes, leaving a 50 byte freed area, and the 10 bytes are later freed, will it result in 2 memory areas tracked by MQX, one 10 bytes and the other 50 bytes, or will it detect that they are adjacent/contiguous in memory and combine them back into 60 byte area?
4. _mem_copy vs memcpy : I do not believe we have overridden the C library memcpy function to use the MQX _mem_cpy function.  What problems will this create?  Do the MQX memory functions handle size differently to accommodate for alignment or fit, whereas the C library might not?
5. Is there a way to get insight into how fragmented the memory is becoming over time? 

Suggestions on best ways to accommodate the need for a lot of dynamic memory use across functional areas of the code and across threads?  Have considered memory pools / blocks, msq queues - not sure which pros & cons of each will impact things the most.

[soledad]

Hi John,

In addition, Freescale MQX uses its own memory allocation system, which doesn't rely on heap.

The default MQX memory pool, where MQX kernel allocates system memory, is between __KERNEL_DATA_START and __KERNEL_DATA_END (symbols generated during bsp linkage).

It allocates memory using function _mem_alloc_system().

On top of this, in bsp we define malloc() to call _mem_alloc_system().

If you want the MQX kernel to call Thumb library malloc, then you have to define _mem_alloc_system* in preprocessor to be malloc #define _mem_alloc_system malloc and so on. And then disable bsp "duplicate" definition of malloc (either by link order or be removing the duplcate definition from the sources).

If you want to use Thumb library allocation system, you can, but it is your responsibility to give the CodeWarrior necessary linker generated files and so on, whatever the Thumb library allocation system relies on.

You lose TAD, kernel logging and other MQX features related to memory allocation.

Have a great day,
Sol

-----------------------------------------------------------------------------------------------------------------------
Note: If this post answers your question, please click the Correct Answer button. Thank you!
-----------------------------------------------------------------------------------------------------------------------

[markschuck · Accepted solution]

1) seems to be limited to the amount of RAM you have on your system (allocations will fail only if you run out of ram, or if it can't find a block big enough)

2) http://cache.freescale.com/files/32bit/doc/user_guide/MQX_User_Guide.pdf?&Parent_nodeId=&Parent_page...

as long as you don't change MQX_MEMORY_FREE_LIST_SORTED from the default '1' then it looks like it will find the best fit because the list is maintained in sorted order.

3) yes

4)  you should be fine either way, i normally use the C library version just in case

5) i don't know what IDE you use, but the MQX plugin for IAR is excellent, it shows details on memory allocations from the various memory pools/blocks and message queue (and just about any other resource you can allocate/initialize or block on).

fyi: I've noticed that many kinetis parts have 2 banks of sram and that mqx will prevent any allocations from spanning both banks (ex: k60 with 2 32 kB banks, mqx allocates 16 bytes at the end of the first bank which can cause problems depending on your specific use case, usually an issue when you need multiple large allocations.

[jpfitz]

Thank you for the answers!  Reading the description in the MQX User Guide MQX_MEMORY_FREE_LIST_SORTED, sounds like it is sorting by physical address, not size.  Therefore I would think it will do a "first fit" not "best fit"

>>

MQX_MEMORY_FREE_LIST_SORTED

Default is one.

One: MQX sorts the freelist of memory blocks by address. This reduces memory

fragmentation, but increases the time MQX takes to free memory.

<<